On Tue, May 6, 2025 at 1:55 AM Vlastimil Babka <vbabka@xxxxxxx> wrote: > > On 5/1/25 05:27, Alexei Starovoitov wrote: > > From: Alexei Starovoitov <ast@xxxxxxxxxx> > > > > Allow __GFP_ACCOUNT and __GFP_COMP flags to be specified when calling > > alloc_pages_nolock(), since upcoming reentrant alloc_slab_page() needs > > to allocate __GFP_COMP pages while BPF infra needs __GFP_ACCOUNT. > > > > Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx> > > I would rather see __GFP_COMP implied as we don't need any new users of > high-order allocations that are not compound, and alloc_pages_nolock() is a > new API. > > IIRC it would be more important in some early version where > free_pages_nolock() was different, now it just calls to ___free_pages() > which has the tricky code for freeing those non-compound allocations. > > But still, I'd really recommend that approach. Note __GFP_COMP is simply > ignored for order-0 so no need to filter it out. All makes sense to me. Will make __GFP_COMP a default for alloca_pages_nolock().
